API Calls¶

This is the documentation of how API works in this system.

Setting Up¶

Create the conda environment and download all the dependencies by:

conda env create

conda activate soo

Training S.O.O.¶

This is how to train S.O.O. on your local computer with CPU. All configurrations are specified in the config files.

1. All the configurations are in the config python file

2. We established a few training script for training S.O.O.

   • Training a plain PPO with customized hyperparemetr specified in the configs:

     python soo/train_ppo.py
     

   • Training just S.O.O. with either with or without the pre-trained core (specify in the configs):

     python soo/train_soo.py
     

   • Training just the transfered soo core with intention data specified in the config path:

     python soo/train_supp.py
     

   • Training the full pipeline all at once sequentially:

     python soo/train_sequential.py
     

3. We have created numerous testing functions and visualization function, including:

   • Testing functions on specified environmental conditions, which takes in configs in the config file as well:

     python soo/test.py
     

   • Visualization of the latent representation of gait cycle, depending on either ppo or soo is used, the code is different:

     python soo/vis/ppo_vis.py
     

     and

     python soo/vis/soo_vis.py
     

   • Still working on evaluation pipeline of full continual learning problem.

Still working on implementing other base comparison algorithms (model-based, sac, mompo, ...)